Repository navigation
generation dtype docs - #12
Merged
Merged
Conversation
decitre
commented
Nov 4, 2026
decitre
commented
Owner
- Document that the fastest generation dtype depends on the backend
- Add a quadrant chart for choosing a generation dtype
Measured 32-token greedy generation for SmolLM2-135M, SmolLM2-360M and TinyLlama-1.1B in Chrome, Safari and Firefox. On WebGPU, q4 is fastest (10 to 15x faster than q8, which is the slowest WebGPU choice despite being first in the default fallback order). On WASM it is the reverse: 4-bit kernels are 4 to 8x slower than q8 and 7 to 11x slower than fp16, the same in all three browsers. A single model file over 2 GB cannot load in a browser, so large models need q8 on WASM. Add a generation dtype section with the measurements to docs/MODELS.md, stop recommending q4 for SmolLM2-360M without naming the backend, point to it from USAGE.md, and note in the WebGPU notebook that most of its WASM time comes from slow 4-bit kernels (WASM fp16 takes about 0.6 s). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Plot tokens/s against dtype for each model and backend, split into quadrants by download size (4/8-bit vs 16/32-bit) and by 10 tokens/s (about twice reading speed), so the q4-on-WebGPU and fp16/q8-on-WASM choices read at a glance. Light and dark renderings are embedded with <picture>. The script declares matplotlib inline (PEP 723), so `uv run scripts/plot_generation_dtypes.py` regenerates both without a new project dependency.
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #12 +/- ##
=======================================
Coverage 82.96% 82.96%
=======================================
Files 8 8
Lines 2472 2472
=======================================
Hits 2051 2051
Misses 421 421
Flags with carried forward coverage won't be shown. Click here to find out more. 🚀 New features to boost your workflow:
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.